batching.md

Event batching

⚠️ Warning: Technical preview

NOTE: Event batching is supported only for:

• runtimes:
  • python
• trigger kinds:
  • http

The event batching feature allows passing a batch of events, instead of a single event, to a function handler. The function invocation remains the same as in the usual flow. The processor keeps each invocation (or event) until a batch is ready to be sent. The batched event is sent to the runtime when either the batch is full or the batching timeout has passed. Both batchSize and timeout are configured per trigger.

Example of trigger configuration with batching enabled:

  triggers:
    http:
      kind: http
      name: http
      batch:
        mode: enable
        batchSize: 10
        timeout: 1s
      maxWorkers: 10

A crucial difference from a single event flow lies in the necessity to manage responses for each event in a batch. Every event in a batch has an event_id field, which should be used in the response when defining the corresponding response for each single event in a batched event. The processor uses response.event_id to determine the correct target for sending each individual response. If response.event_id isn't set, the user won't receive a response.

Handler example:

import nuclio_sdk

def handler(context, batch: list[nuclio_sdk.Event]):
    context.logger.info_with('Got batched event!')
    batched_response = []
    response = process_batch(batch)
    for item in batch:
        event_id = item.id
        batched_response.append(nuclio_sdk.Response(
            body=response,
            headers={},
            content_type="text/plain",
            status_code=200,
            event_id=event_id,
        ))
    return batched_response

def process_batch(event: list[nuclio_sdk.Event]):
    return "Hello"